Editorial
The rest of this site is a comparison directory: scores, prices, benchmarks. These are the pieces where we write in our own voice — the reasoning behind our methodology, and what actually happens when the AI infrastructure we run ourselves breaks in production. Nothing here is generated from the model catalog.
Editorial · Methodology
A benchmark leaderboard can tell you a model wins. It can't tell you whether the provider will still sell it to you next quarter, or whether the win matters for what you're actually building. Here's the judgment layer we add on top, and the mistake that taught us to add it.
Editorial · Field Notes
This site's own 'Ask AI' feature has been taken down three separate times by the exact kind of provider risk that never shows up in a benchmark table. Model IDs get retired overnight. Free tiers hit invisible ceilings. Here's the real timeline, in real numbers.
Editorial · Opinion
A score out of 100 is real, and it's also a snapshot. It can't tell you whether the model will still be sold to you in three months, whether its rate limits survive contact with real traffic, or whether the leaderboard you're reading was actually re-checked after the last launch. Four things we've learned the hard way that no benchmark table captures.